Method: IO.select
- Defined in:
- io.c
permalink .select(read_ios, write_ios = [], error_ios = [], timeout = nil) ⇒ Array?
Invokes system call select(2), which monitors multiple file descriptors, waiting until one or more of the file descriptors becomes ready for some class of I/O operation.
Not implemented on all platforms.
Each of the arguments read_ios
, write_ios
, and error_ios
is an array of IO objects.
Argument timeout
is a numeric value (such as integer or float) timeout interval in seconds.
The method monitors the IO objects given in all three arrays, waiting for some to be ready; returns a 3-element array whose elements are:
-
An array of the objects in
read_ios
that are ready for reading. -
An array of the objects in
write_ios
that are ready for writing. -
An array of the objects in
error_ios
have pending exceptions.
If no object becomes ready within the given timeout
, nil
is returned.
IO.select peeks the buffer of IO objects for testing readability. If the IO buffer is not empty, IO.select immediately notifies readability. This “peek” only happens for IO objects. It does not happen for IO-like objects such as OpenSSL::SSL::SSLSocket.
The best way to use IO.select is invoking it after non-blocking methods such as #read_nonblock, #write_nonblock, etc. The methods raise an exception which is extended by IO::WaitReadable or IO::WaitWritable. The modules notify how the caller should wait with IO.select. If IO::WaitReadable is raised, the caller should wait for reading. If IO::WaitWritable is raised, the caller should wait for writing.
So, blocking read (#readpartial) can be emulated using #read_nonblock and IO.select as follows:
begin
result = io_like.read_nonblock(maxlen)
rescue IO::WaitReadable
IO.select([io_like])
retry
rescue IO::WaitWritable
IO.select(nil, [io_like])
retry
end
Especially, the combination of non-blocking methods and IO.select is preferred for IO like objects such as OpenSSL::SSL::SSLSocket. It has #to_io method to return underlying IO object. IO.select calls #to_io to obtain the file descriptor to wait.
This means that readability notified by IO.select doesn’t mean readability from OpenSSL::SSL::SSLSocket object.
The most likely situation is that OpenSSL::SSL::SSLSocket buffers some data. IO.select doesn’t see the buffer. So IO.select can block when OpenSSL::SSL::SSLSocket#readpartial doesn’t block.
However, several more complicated situations exist.
SSL is a protocol which is sequence of records. The record consists of multiple bytes. So, the remote side of SSL sends a partial record, IO.select notifies readability but OpenSSL::SSL::SSLSocket cannot decrypt a byte and OpenSSL::SSL::SSLSocket#readpartial will block.
Also, the remote side can request SSL renegotiation which forces the local SSL engine to write some data. This means OpenSSL::SSL::SSLSocket#readpartial may invoke #write system call and it can block. In such a situation, OpenSSL::SSL::SSLSocket#read_nonblock raises IO::WaitWritable instead of blocking. So, the caller should wait for ready for writability as above example.
The combination of non-blocking methods and IO.select is also useful for streams such as tty, pipe socket socket when multiple processes read from a stream.
Finally, Linux kernel developers don’t guarantee that readability of select(2) means readability of following read(2) even for a single process; see select(2)
Invoking IO.select before IO#readpartial works well as usual. However it is not the best way to use IO.select.
The writability notified by select(2) doesn’t show how many bytes are writable. IO#write method blocks until given whole string is written. So, IO#write(two or more bytes)
can block after writability is notified by IO.select. IO#write_nonblock is required to avoid the blocking.
Blocking write (#write) can be emulated using #write_nonblock and IO.select as follows: IO::WaitReadable should also be rescued for SSL renegotiation in OpenSSL::SSL::SSLSocket.
while 0 < string.bytesize
begin
written = io_like.write_nonblock(string)
rescue IO::WaitReadable
IO.select([io_like])
retry
rescue IO::WaitWritable
IO.select(nil, [io_like])
retry
end
string = string.byteslice(written..-1)
end
Example:
rp, wp = IO.pipe
mesg = "ping "
100.times {
# IO.select follows IO#read. Not the best way to use IO.select.
rs, ws, = IO.select([rp], [wp])
if r = rs[0]
ret = r.read(5)
print ret
case ret
when /ping/
mesg = "pong\n"
when /pong/
mesg = "ping "
end
end
if w = ws[0]
w.write(mesg)
end
}
Output:
ping pong
ping pong
ping pong
(snipped)
ping
11116 11117 11118 11119 11120 11121 11122 11123 11124 11125 11126 11127 11128 11129 11130 11131 11132 11133 11134 11135 11136 11137 11138 11139 11140 11141 11142 11143 11144 |
# File 'io.c', line 11116
static VALUE
rb_f_select(int argc, VALUE *argv, VALUE obj)
{
VALUE scheduler = rb_fiber_scheduler_current();
if (scheduler != Qnil) {
// It's optionally supported.
VALUE result = rb_fiber_scheduler_io_selectv(scheduler, argc, argv);
if (!UNDEF_P(result)) return result;
}
VALUE timeout;
struct select_args args;
struct timeval timerec;
int i;
rb_scan_args(argc, argv, "13", &args.read, &args.write, &args.except, &timeout);
if (NIL_P(timeout)) {
args.timeout = 0;
}
else {
timerec = rb_time_interval(timeout);
args.timeout = &timerec;
}
for (i = 0; i < numberof(args.fdsets); ++i)
rb_fd_init(&args.fdsets[i]);
return rb_ensure(select_call, (VALUE)&args, select_end, (VALUE)&args);
}
|